Abstract: Image segmentation is an important process in computer vision. Most of the currently popular methods tracking the object edge depend on pixel gray value. Because of the noise in images, the pixel resolving and the precision of the detection method itself, the detected contours are far from smoothness and with many tiny zigzags, which will affect the subsequent processing such as image recognition, image feature detection et al. This paper post forwards a novel method to get smooth contour in image segmentation process. Firstly, adopt dynamic programming algorithm to process the object image and to get a global optimal boundary, then smoothing and fitting the boundary by adaptive B-Spline, which adjusts the control points depending on the contour curvature. Practical numerical experimental results show that this method has stronger edge detection ability and produced smoother contour curve than other methods and without losing edge feature at the same time.